TURING v3.0.0 — Retrieval-Augmented Generation

TURING v3.0.0

TURING v3 introduces Retrieval-Augmented Generation (RAG), extending the agent with semantic document retrieval using embeddings and vector search.

What's New

• Added document chunking
• Added Ollama-based embeddings
• Added cosine similarity search
• Added in-memory vector store
• Added top-k document retrieval
• Added RAG context building
• Added document indexing pipeline
• Integrated RAG with the existing agent architecture
• Added support for semantic retrieval from indexed documents

RAG Pipeline

Document → Chunking → Embeddings → Vector Store → Similarity Search → Retrieved Context → LLM

Architecture

The RAG system is implemented as a modular pipeline inside the RAG/ package, keeping document processing, embeddings, storage, similarity calculation, and retrieval separated.

Testing

• Full automated test suite included
• Existing TURING functionality remains intact
• RAG components integrated with the existing architecture

Project Status

TURING v3.0.0 is a stable release focused on adding retrieval capabilities while keeping the system lightweight, local, and modular.

Built with Python, Ollama, Qwen 2.5, Ollama Embeddings, SQLite, Tavily, and pytest.

What's Changed

• chore: finalize project and documentation by @zermello in #5

New Contributors

• @zermello made their first contribution in #5

Full Changelog: v2.0.0...v3.0.0

TURING v2 — Tooling & Document Intelligence

TURING v2.0.0 — Tooling & Document Intelligence

TURING v2 is a major milestone focused on expanding the agent’s tool ecosystem, persistent memory, document processing, and overall reliability.

What’s included

🧠 LLM & Conversation

• Local LLM integration with Ollama
• Conversational context
• Tool calling
• Multiple tool calls

🛠️ Tool System

• Dynamic tool registry
• Calculator tool
• Weather tool
• Time tool
• Web search tool
• Document reader tool

💾 Memory

• Persistent SQLite memory
• Memory storage and retrieval through the agent

📄 Document Processing

TURING can now process multiple document formats:

• .txt
• .md
• .json
• .csv
• .pdf
• .docx

The document registry automatically selects the appropriate reader based on the file extension.

🧪 Reliability & Testing

• Automated testing with pytest
• Error handling
• Structured logging
• Tool execution testing
• Document reader tests

Architecture

TURING now follows a more modular architecture where tools are registered centrally and can be selected and executed by the agent.

User
↓
TURING
↓
Intent / Tool Selection
↓
Dynamic Tool Registry
↓
Tool Execution
↓
Result
↓
LLM
↓
Response

Document processing extends this architecture:

Document
↓
Document Registry
↓
File Type Detection
↓
Specialized Reader
↓
Extracted Content
↓
TURING

What’s next

The next major milestone is TURING v3, focused on Retrieval-Augmented Generation (RAG).

Planned RAG pipeline:

Documents
↓
Document Processing
↓
Chunking
↓
Embeddings
↓
Vector Storage
↓
Retrieval
↓
Relevant Context
↓
LLM
↓
Grounded Response

TURING v3 will build on the document-processing foundation introduced in this release.

Status

v2.0.0 — Released

Thanks for checking out TURING. 🚀

Full Changelog: v0.1.0...v2.0.0

TURING v0.1.0 — First Stable Milestone

What’s included

• Calculator tool
• Weather tool
• Time tool
• Persistent SQLite memory
• Multi-tool routing
• Multiple tool-call support
• Basic error handling
• Agent logging
• Local Ollama model support

Status

First stable development milestone of TURING.

This release establishes the core foundation for future improvements in tool architecture, memory, reasoning, and agent capabilities.